Bugfix/fix nvidia docker v2 (redux)

[esteve]

(Copied from #1167)

Status

PRODUCTION, safe for the "optional" todos mentioned below.

Description

Adding nvidia docker v2 support as discussed in issue autowarefoundation/autoware_ai#93 .

Todos

I'd like to leave these up to the Autoware maintainers.

• I think these new images should become the standard (as Autoware assumes or even needs an nvidia GPU to run on, right?) You can consider renaming current images to something like "-legacy", rather than suffixing these new images with "-nvidia".
• Documentation could be checked on all old references to the old nvidia-docker run command (which is docker run --runtime nvidia these days).

Steps to Test or Reproduce

See autowarefoundation/autoware_ai#93.

Make sure to install nvidia-docker v2 and see if the old image works for your (should not work. Opening RVIZ would crash for example). Then try again with the new image and see if it works.

[esteve]

This is still a work in progress. I simply rebased #1167 and pushed it here.

[asimonov-hitachi]

I can confirm this solves the OpenGL issues with opening rviz

[gbiggs]

@esteve Is this PR still valid?

[MTDzi]

@esteve Is this PR still valid?

Same question here.

[JWhitleyWork]

@esteve ping

[JWhitleyWork]

@gbiggs / @MTDzi - Based on

autoware/docker/generic/run.sh

Line 123 in e7c12b0

RUNTIME="--runtime=nvidia"

it looks like at least master supports the --runtime=nvidia option which is only available with nvidia_docker2 (nvidia_docker used the actual nvidia-docker command instead). I just tested it with bash autoware/docker/generic/run.sh and it appears to work correctly on my 18.04 host. The results from nvidia-smi look correct.

[JWhitleyWork]

#1946 fixed this particular issue. I am pretty positive this PR can be closed.

[gbiggs]

Awesome! Thanks for letting us know.